www.nature.

com/mp Molecular Psychiatry

EXPERT REVIEW OPEN

Computational psychiatry: from synapses to sentience

1✉
Karl Friston

© The Author(s) 2022

This review considers computational psychiatry from a particular viewpoint: namely, a commitment to explaining psychopathology
in terms of pathophysiology. It rests on the notion of a generative model as underwriting (i) sentient processing in the brain, and (ii)
the scientific process in psychiatry. The story starts with a view of the brain—from cognitive and computational neuroscience—as
an organ of inference and prediction. This offers a formal description of neuronal message passing, distributed processing and
belief propagation in neuronal networks; and how certain kinds of dysconnection lead to aberrant belief updating and false
inference. The dysconnections in question can be read as a pernicious synaptopathy that fits comfortably with formal notions of
how we—or our brains—encode uncertainty or its complement, precision. It then considers how the ensuing process theories are
tested empirically, with an emphasis on the computational modelling of neuronal circuits and synaptic gain control that mediates
attentional set, active inference, learning and planning. The opportunities afforded by this sort of modelling are considered in light
of in silico experiments; namely, computational neuropsychology, computational phenotyping and the promises of a
computational nosology for psychiatry. The resulting survey of computational approaches is not scholarly or exhaustive. Rather, its
1234567890();,:

aim is to review a theoretical narrative that is emerging across subdisciplines within psychiatry and empirical scales of investigation.
These range from epilepsy research to neurodegenerative disorders; from post-traumatic stress disorder to the management of
chronic pain, from schizophrenia to functional medical symptoms.

Molecular Psychiatry (2023) 28:256–268; https://doi.org/10.1038/s41380-022-01743-z

INTRODUCTION theory or hypothesis-led approaches, where hypotheses are

The inception of computational psychiatry—a decade ago—
reflected a consensus that a formal, crisp and clear approach to
psychiatry could be furnished by computational advances in data
modelling, neuroscience and machine learning [1–5]. There are
now annual workshops in computational psychiatry and an
eponymous journal. However, introducing computational psychia-
try was a risky move: most editors know that that pre-pending any
journal title with “Computational” is likely to drop its impact factor
by at least one point. So, has computational psychiatry delivered?
And where is it going?
Thirty years ago, state-of-the-art computational techniques
meant the application of factor analysis to discern patterns of
symptoms and signs and schizophrenia—e.g. [6]—or, at least for
me, the use of canonical correlation analysis to describe patterns
of cerebral blood flow in brain imaging [7]. Nowadays, things are
very different. I spend my time discussing the finer details of
modelling synaptic efficacy with researchers, whose expertise
spans magnetoencephalography, functional genomics, optoge-
netics and cell cultures. I sit there—pretending to know all the
acronyms—and marvel at how a common, computational,
narrative can bring molecular biologists and clinical neuroscien-
tists to the same table. This review considers some of the
foundational tenets of this narrative, illustrating the ‘work in
progress’ with a few select examples.
Our focus will be on computational ideas and procedures that
have endured, sometimes for centuries, and are currently [re]
emerging across theoretical neurobiology. This limits us to a
articulated as a generative model, for which empirical evidence
can be sought. This demarcates computational approaches with
‘strong’ explanatory power from ‘weak’, descriptive or theory-free
approaches. Many ‘weak’ approaches have come and gone,
leaving an underwhelming legacy. Examples of weak approaches
include behaviourism [8]; namely, a description of behaviour that
is reinforced by an outcome that is defined (tautologically) by its
ability to reinforce behavior [Clearly, behaviourism is not theory
free. I am using ‘behaviourism’ here as a euphemism for simply
collecting and summarising dependencies in observable beha-
viour (e.g., with an autoregression or Rescorla-Wagner model)].
Strong alternatives rest upon the explanatory power of active
inference and learning [9], that emphasise beliefs, uncertainty,
preferences and curiosity [10–14]. In imaging neuroscience a weak
example would be correlation patterns in functional magnetic
resonance imaging (fMRI) or electrophysiology, a.k.a., functional
connectivity [15]; where its strong counterpart would be the
directed connectivity in neuronal circuits that cause the correla-
tions, a.k.a., effective connectivity [16–22]. In the data sciences,
weak computational approaches rest on ‘big data’ and machine
(e.g., deep) learning procedures that preclude generative or world
models and, as a consequence, any mechanistic interpretability or
explainability [20, 23, 24]. The strong complement, in machine
learning, rests on generative modelling; e.g., generative adversar-
ial networks and automatic variational inference [25–27]. [The
bright line between strong (hypothesis-led) and weak (data-led)
approaches is, of course, fragile. For example, weak approaches

Wellcome Centre for Human Neuroimaging, Institute of Neurology, University College London, London WC1N 3AR, UK. ✉email: k.friston@ucl.ac.uk
1

Received: 27 March 2022 Revised: 8 August 2022 Accepted: 11 August 2022

Published online: 2 September 2022

are necessary to furnish evidence for strong approaches (e.g.,
genome-wide association studies GWAS that point to particular
synaptopathies), while hypotheses are necessary to make the best
sense of data (e.g., enrichment analysis in prioritizing variants from
GWAS)].
The common theme that distinguishes weak from strong
approaches is reference to a generative model: namely, a
hypothesis about causal mechanisms, specified formally as a joint
probability distribution over some (unobservable, latent or
hidden) causes and their (observable, measurable) consequences.
What follows is, at every level, undergirded by a generative model
that places (strong) computational psychiatry—and perhaps our
own sentience—firmly in the realm of mechanistic, evidenced-
based science; in the sense, that one can compare the evidence
for one generative model, relative to another.
OVERVIEW
The first section sets the scene for a mechanistic formulation of
psychopathology under current ideas about the brain as an organ
of inference, prediction and planning. These ideas can be
articulated in terms of the computational processes that
accompany belief updating in the Bayesian brain [28–32]. The
second section turns to psychopathology—and the underlying
pathophysiology—by asking how aberrant belief updating and
predictive processing might manifest? In brief, the conclusion is
that many psychiatric and neurological disorders can be framed in
terms of a synaptopathy that confounds the encoding of
uncertainty or precision in a (Bayesian) brain [33–35].
The third section considers the imperatives for computational
modelling of empirical data that inherit from the previous sections.
This leads to a discussion of how to assess the modulation of
synaptic efficacy in terms of changes in effective connectivity, with
an emphasis on dynamic causal modelling in psychiatry [20, 36, 37].
The final sections deal with computational phenotyping; namely,
using computational or in silico models of belief updating—and
psychophysical or behavioural concomitants—to explain the
responses of a given subject. They briefly consider computational
nosology, by asking if there is a generative model of pathophysiol-
ogy and psychopathology that can be used to best explain the
trajectories of individual patients using high-density longitudinal
data. Finally, the opportunity afforded by a formal and mechanistic
understanding of functional brain architectures is exemplified by
computational neuropsychology; namely, performing in silico lesion
or pharmacological experiments.
THE BAYESIAN BRAIN, PRECISION ENGINEERING AND ACTIVE
INFERENCE
The narrative starts with the simple premise that if psychology—
read as belief updating in the brain—can be cast as a
computational process of inference, it follows that psychopathol-
ogy just is false inference. False inference is meant in the usual
sense of false positives (i.e., type I errors); namely, inferring
something is there when it is not. Cardinal examples here include
hallucinations, delusions and other features of reality distortion
seen in psychosis. False negatives (i.e., type II errors) mean
inferring something is not there when it is; for example,
dissociative disorders, neglect syndromes, derealisation phenom-
ena, et cetera. Indeed, when one thinks about psychiatric and
neurological disorders, most can be framed as false inference:
ranging from dysmorphophobia in eating disorders, through to
delusional systems in paranoid schizophrenia; from phobias
through to Parkinson’s disease. It may seem odd to include
Parkinson’s disease; however, the pathognomonic bradykinesia—
and failure to initiate movement—can, on one reading, be seen as
a failure to realise motor planning as inference (see below). So,
what licences the assumption that psychology is inference?
Molecular Psychiatry (2023) 28:256 – 268
K. Friston
257
[Clearly, asserting that all of psychopathology is ‘false inference’
requires some qualification. The basic argument—that licences
this assertion—is that any symptom (or sign) of psychopathology
must, at some level, entail aberrant belief updating. The kind of
beliefs here are Bayesian beliefs; namely, subpersonal or proposi-
tional representations that possess the attribute of uncertainty,
and are conditional in a well-specified sense. Under this reading of
beliefs, sentient behaviour—and beliefs about that behaviour—
can then be cast as inference (e.g., perceptual inference, planning
as inference, attribution of agency, emotional recognition, et
cetera). Reducing everything to inference then provides a
mechanics to understand neuronal dynamics and development,
in terms of belief updating and the requisite message in the
brain].
The foundations of this account of sentient behaviour can be
traced back to the days of Plato and were most clearly articulated
in the 19th-century by Helmholtz as unconscious inference
[38, 39]—ideas that are reminiscent of Kantian philosophy. These
ideas endured through the 20th-century in several flavours; for
example, analysis by synthesis [40, 41], epistemological automata
[42], perception as hypothesis testing [43, 44] and, in machine
learning, the Helmholtz machine [45]. The inference narrative
supervened at the turn of the century, with a resurgence of
interest in enactivist approaches [46–51] that now predominate in
the cognitive and systems neurosciences, in the form of things like
predictive processing and active inference [52–60].
Active inference can be read as an enactive version of the
Bayesian brain hypothesis [61, 62] that subsumes sentience
(perceptual inference) and behaviour by treating control and
planning as inference [28–31]. So, what is inference? In this setting,
inference just refers to a process that maximises the evidence for
some (generative) model or hypothesis about the causes of
(sensory) data. Model evidence is also known as marginal likelihood;
namely, the likelihood of some data, under a model of how those
data were generated. Maximising the evidence for our own
generative model is sometimes called self-evidencing [63]. In brief,
active inference casts the brain as a fantastic organ: a generator of
fantasies, hypotheses and predictions that are tested against
sensory evidence. One might ask how this account of sentient
behaviour speaks to neuroanatomy, and the functional architec-
tures implied by neurophysiology.
The answer is relatively straightforward. Given a generative
model, there are well described belief updating or propagation
schemes that specify the requisite message passing that must, in
some form, be implemented by neuronal networks. For generative
models based upon continuous states of the world, these schemes
are known as Bayesian filters or predictive coding [56, 64–67]. In
generative models of discrete states (e.g., “I am in the kitchen”, as
opposed to “I am at these continuous GPS coordinates”) the
message passing schemes are variously known as belief propaga-
tion or variational message passing [68–71]. All of these schemes
can be cast as a gradient ascent on model evidence or marginal
likelihood [72]. In short, neuronal dynamics just are a process of
inference. See Fig. 1 for a schematic description of predictive
coding.
Crucially, the gradients that subtend neuronal dynamics—and
consequent belief updating—can always be formulated as a
prediction error [This may sound like a sweeping statement;
however, any random dynamical system (including neuronal
dynamics) can be cast as a gradient flow, where the gradients can
always be expressed as a difference in log probabilities, which can
be read as prediction errors of one kind or another] [70]. In other
words, the divergence between predictions of sensory input and
the observed sensations. In predictive coding schemes, it is thought
that prediction errors are represented explicitly: e.g., by superficial
pyramidal cells in the upper layers of the cortex [66, 73–76].
This leads to a picture of hierarchical inference in the brain as
the reciprocal message passing between the levels of a cortical
 K. Friston
258

Fig. 1 This figure summarizes hierarchical message passing in predictive coding. Predictive coding is a scheme for generative models
based upon continuous states. In these schemes, neuronal activity encodes expectations about the causes of sensory input that are
continually updated to minimize prediction error. Prediction error is the difference between (ascending) sensory input and (descending)
predictions of that input. This minimization rests upon recurrent neuronal interactions among different levels of a cortical hierarchy. It is
thought that superficial pyramidal cells (red triangles) compare the expectations (at each hierarchical level) with top-down predictions from
deep pyramidal cells (black triangles) of higher levels: see [66] for a review. Left panel: this schematic shows a simple cortical hierarchy with
ascending prediction errors and descending predictions. It includes neuromodulatory gating or gain control (teal) of superficial pyramidal
cells that determines their relative influence on deep pyramidal cells encoding expectations. Right panel: this provides a schematic example in
the visual system: it shows the putative cells of origin of ascending or forward connections that convey prediction errors (red arrows) and
descending or backward connections (black arrows) that furnish predictions. The prediction errors are weighted by their expected precision,
which have been associated with projections from the pulvinar. In this example, the frontal eye fields send predictions to primary visual
cortex, which projects to the lateral geniculate body. However, the frontal eye fields also send proprioceptive predictions to pontine nuclei,
which are passed to the oculomotor system to induce movement, through classical oculomotor reflexes. This means that saccadic eye
movements, for example, realise top-down proprioceptive predictions that are deeply informed by hierarchical processing of data from all
sensory modalities. Similar schemes can be elaborated for discrete state space generative models, responsible for planning and motor control
—or allostasis and autonomic function: see [54] for equivalent schematics and [70] for mixtures of continuous and discrete models.

hierarchy; in which prediction errors ascend from lower to higher
levels to drive changes in neuronal populations encoding states of
affairs in the world. These populations (e.g., deep pyramidal cells
in lower layers of the cortex) then supply a counter stream of
descending predictions that resolve or cancel prediction errors at
lower levels [66, 75, 77]: e.g., by targeting inhibitory interneurons
that are coupled to the superficial pyramidal cells broadcasting
prediction errors [78, 79]. This architecture also plays out under
discrete generative models and has become something of a
workhorse for understanding recurrent message passing in
cortical and subcortical hierarchies.
THE IMPORTANCE OF BEING PRECISE
Prediction errors (i.e., the gradients that drive belief updating) can
be regarded as carrying the newsworthy information at any given
hierarchical level to the level above. However, this is not the
complete story. Higher levels have to select which prediction
errors to listen to; much in the same way that we select our most
trustworthy news channels or sources of information. This
selection rests upon predictions of predictability or precision.
Predicting or estimating precision is a universal requisite for
making sense of data: from estimating the signal-to-noise ratio in
some sensory apparatus, through to estimating standard error
when making inferences via some Neyman-Pearson statistic.
Affording certain prediction errors greater precision increases
their influence on belief updating and has all the hallmarks of
attentional selection [80–85]. Physiologically, this simply entails an
increase in the excitability or postsynaptic gain of neuronal
populations broadcasting prediction errors. On this view, there is
an intimate relationship between attention and the modulation of
synaptic efficacy by classical neuromodulators and nonlinear
postsynaptic responses responsible for mediating the exchange
between fast-spiking inhibitory interneurons and (superficial)
pyramidal cells [78, 79, 81, 86–91]. Please see Table 1 and [92]
for a fuller discussion of neuromodulators and the representation
of precision. Figures 2, 3 illustrate the neuronal circuitry implicated
in precision or gain control, with a focus on canonical microcircuits
and inhibitory interneurons, respectively.
One crucial aspect of this precision engineering is that it
underwrites our ability to filter out—or ignore—certain prediction
errors when they are deemed imprecise. A key example is the
attenuation of sensory prediction errors that report the con-
sequences of movement [93–98]. If we could not ignore the
proprioceptive and somatosensory afferents—supplying evidence
that we are not moving—then any beliefs about intended or
predicted movement would be revised immediately, and we
would not be able to initiate movement. This mandates a transient
suspension of sensory precision during active sensing: c.f.,
saccadic suppression of optic flow signals during saccadic eye
movements [99]. It also provides a glimpse of how one might
explain Parkinson’s disease in terms of dopaminergic failures of
sensory attenuation [100]. More generally, it foregrounds the role
of precision in orchestrating the perception action cycles [101]
that underwrite active sensing [102]. For example, sensory
attenuation suggests a particular scheduling of action and
perception: although both work hand-in-hand to resolve predic-
tion errors, this resolution may involve a fast (~4 Hz) alternation
between sampling the world (e.g., visual palpation with saccades)
and belief updating after each sample.

Molecular Psychiatry (2023) 28:256 – 268

 K. Friston
259
Table 1. The pharmacology of precision. Please see [92] for details.
Neurotransmitter system Functional role Neuroanatomy
Cholinergic Encoding the precision of outcomes given hidden states: (c.f., Nucleus basalis of Meynert
Attention and expected uncertainty [218, 219])
Noradrenergic Encoding the precision of state transitions (c.f., Volatility and Locus coeruleus
unexpected uncertainty [220])
Dopaminergic Encoding the precision of beliefs about policies (c.f. Action selection Substantia nigra pars compacta, Ventral
[221]) tegmental area

Fig. 2 Predicting precision. This schematic is a more detailed version of Fig. 1 that includes putative laminar-specific connections that are
consistent with the precision-based predictive coding described in [176]. This architecture is based upon [66] and conforms roughly to the
known neuroanatomy and physiology of canonical microcircuits in the visual system and the laminar specificity of extrinsic connections. The
key aspect of this figure is the inclusion of deep pyramidal cells encoding the amplitude of (squared or unsigned) prediction error that inform
posterior expectations about precision in the (matrix cells) of the pulvinar. These cells reciprocate descending projections to modulate the
gain of superficial pyramidal cells in cortex. Forward connections are in red, and descending (backward) connections are in black. First-order
prediction errors are shown as full red lines and second-order (unsigned) prediction errors are shown as broken red lines. This schematic
ignores inhibitory interneurons that mediate some inferences vicariously. See subsequent figure.

In summary, the narrative so far is that psychopathology
represents false inference or aberrant belief updating, under a
view of the brain as a statistical organ, generating predictions and
revising its (subpersonal Bayesian) beliefs on the basis of
prediction errors. Crucially, these predictions are contextualised
with predictions of precision or predictability that instantiate
attentional or intentional set; allowing the selection of attenuation
of prediction errors via a process of precision weighting. This
precision weighting is nothing more than modulating the gain,
postsynaptic sensitivity or excitability of appropriate neuronal
populations. So, what does this computational architecture offer in
terms of potential targets for pathology?
SYNAPTOPATHY, DYSCONNECTIONS AND FALSE INFERENCE
The narrative returns here to the 19th-century and early formula-
tions of psychiatric disorders such as dementia praecox and
schizophrenia. A common theme of these formulations is a
disintegration or disruption of neuronal processing. From the
perspective of the current narrative, these formulations came in two
flavours. First, Wernicke’s sejunction hypothesis [103] emphasised
disruption to the organs of connection (i.e., white matter tracts in
the brain), positing a form of disconnection syndrome [104]. A
complementary perspective was provided by Bleuler’s notion of
disintegration of the psyche [105], cast in more functional terms
that we might now read as a failure of synaptic integration. These
ideas re-emerged with the advent of functional brain imaging in the
1990s, in the form of disconnection hypotheses, drawing analogies
with things like metachromatic leukodystrophy [106] and the
dysconnection hypothesis [107] that emphasised dysfunctional
synaptic integration; specifically, neuromodulatory failures of
synaptic function [34]. The latter formulation of dysfunctional
connectivity is now arguably the prevalent view for most psychiatric
and neurodegenerative conditions, while variants of the sejunction
hypothesis would be apt for cerebrovascular accidents, space
occupying lesions and other interruptions to white-matter fasciculi:
for example, increases in conduction delays due to cerebral small
vessel disease in white matter fasciculi.
In short, one could neatly summarise the pathophysiology of
many psychiatric and neurological conditions in terms of one or
more forms of synaptopathy [108]. Synaptopathy is taken to mean
any failure of synaptic function due to a variety pathological

Molecular Psychiatry (2023) 28:256 – 268

 K. Friston
260
Interactions between pyramidal cells and inhibitory interneurons

A simplified scheme

Fig. 3 The left panel depicts interactions between (superficial and deep) pyramidal cells with inhibitory interneurons. Inhibitory
interneurons are partitioned into three subtypes (Parvalbumin positive PV, somatostatin SST, and vasoactive intestinal peptide expressing
interneurons, VIP), based upon optogenetic studies [217]. This schematic illustrates the intimate relationship between pyramidal cells and
inhibitory interneurons responsible for excitation-inhibition balance and ensuing excitability in canonical microcircuits. Here, it is assumed
that PV interneurons are reciprocally connected to the pyramidal cells through perisomatic compartments, while SST cells form synapses on
their dendrites. The right panel shows a simplified architecture used in many modelling studies. Here, the recurrent inhibitory (PV/pyramidal
cell) dynamics have been absorbed into an inhibitory recurrent connection, while the SST/VIP interneurons provide (dendritic) inhibitory
drive. This allows one to map models of Interneuron Network Gamma (ING) oscillations onto canonical microcircuits used in dynamic causal
modeling (see next figure). In this setting, the Pyramidal ING (PING) model emphasizes recurrent interactions among PV cells, as modeled by
the inhibitory recurrent connections on superficial pyramidal cells. In contrast, the ING model corresponds to the influence of (SST/VIP)
inhibitory interneurons on pyramidal cells. Please see [81] for further details of how these simplified architectures are used to model
attentional effects on evoked responses.

mechanisms (i.e., formation, structure, metabolism, etc.). For
example, the severe loss of synapse density in human Alzheimer’s
disease, frontotemporal dementia, Parkinson’s disease, Progressive
Supranuclear Palsy, etc. [109].
But what kind of synaptopathy? A coarse-grained overview of the
synaptic theories of schizophrenia suggests that the synaptopathy
in question is of a neuromodulatory sort; implicating classical
neuromodulatory (ascending) neurotransmitter systems [110–113],
GABAergic neurotransmission and NMDA receptors [113–116],
where synchronous interactions between fast spiking inhibitory
interneurons and pyramidal cells may set the excitability of
canonical microcircuits [88, 91, 116–119]. The same story emerges
from the functional genomics of schizophrenia, which point to the
mechanisms that underwrite neuromodulation, synaptic gain
control and ensuing excitation-inhibition balance in neuronal
circuits [120, 121]. For example, a recent genome-wide association
study [122] of 76,755 individuals with schizophrenia highlights the
importance of GRIN2A (which encodes a subunit of the NMDA
receptor), while a recent meta-analysis of whole exomes [123]
emphasises the role of GRIN2A, as well as GRIA3 (which encodes a
subunit of the AMPA receptor). Finally, it is worth noting that nearly
all psychoactive drugs, including psychedelics, target classical
neuromodulatory or NMDA receptors: e.g., [124, 125].
In short, many psychiatric, neurological and neurodegenerative
disorders could be described as arising from pernicious synapto-
pathies that lead to a functional disintegration of neuronal
message passing in cortical and subcortical hierarchies. The
particular synaptopathies in question implicate neuromodulatory
gain control, of the sort required to deploy precision (i.e., selective
attention and sensory attenuation) during inference and planning
in the Bayesian brain.
TOO MUCH OR TOO LITTLE?
This completes the next step in the narrative; namely, the
psychopathology (i.e., false inference) characteristic of psychiatric
disorders may be attributable to aberrant precision control, which
inherits from synaptopathies that confound neuromodulation.
This narrative first emerged in theoretical treatments of halluci-
nosis in synucleinopathies [126–128] and, from a computational
perspective, was developed in schizophrenia [129–133] and
autism research [134–136]. Since that time, it has become difficult
to find a psychiatric condition that has not been considered
through the lens of aberrant precision, in one form or another
[21, 135, 137–161].
A crosscutting theme in many of these accounts is a putative
failure of sensory attenuation [53, 93–95, 98, 100, 159, 162–165]. In
other words, a neuromodulatory failure to attenuate the
postsynaptic gain of neuronal populations reporting the con-
sequence of action (in the motor domain or mediated by
autonomic reflexes). A failure of sensory attenuation implies an
imbalance between the precision afforded sensory prediction
errors and the precision of prediction errors deeper in neuronal
hierarchies that mediate or maintain prior beliefs of a subpersonal
or propositional nature. The ensuing imbalance is often read as a
loss of precision or confidence in prior beliefs, relative to the
sensory evidence at hand. In autism, this imbalance is consistent
with a failure of sensory attenuation and an inability to ignore the
sensorium, which may or may not be associated with the
neuromodulatory role of oxytocin [135, 166–170].
In schizophrenia, the story is a little more involved; in the sense
that a failure of sensory attenuation provides an apt explanation
for resistance to illusions (that normally depend upon precise prior
beliefs) and failures to elicit mismatch oddball responses (because
everything is surprising). Some people then interpret delusional
ideation as the brain’s attempt to make sense of unattenuated
prediction errors: see also [152, 155, 171, 172]. However,
hallucinatory phenomena require a slightly more delicate argu-
ment; usually along the lines of a compensatory increase in prior
precision that could manifest as a form of paradoxical lesion. In
other words, in attempt to override precise sensory prediction
errors, higher levels learn to ignore sensory evidence and—

Molecular Psychiatry (2023) 28:256 – 268


sequestered from the sensorium—elaborate false percepts. See
[22, 133, 141, 158, 173] for related discussion.
There are many interesting instances of this formulation;
ranging from failures of sensory attenuation in autonomic and
interoceptive inference, through depression and emotional
processing [82, 140, 149, 174], to functional medical symptoms
and dissociative phenomena [137, 164]. These computational
formulations are nice because they lend themselves to various
lines of enquiry; ranging from simulating false inference; e.g.,
[149], through the careful analysis of psychophysics and choice
behaviour e.g., [154, 155, 175]; and, crucially, neurophysiological
measurements of synaptic function—to which we now turn.
SYNAPTOPATHY, DYSCONNECTIONS AND DYNAMIC CAUSAL
MODELLING
The narrative so far is that psychopathology is a pernicious form of
false inference that can be attributed to a synaptopathy that
confounds the deployment of precision during perceptual
inference and ensuing action. This narrative is based upon first
principle (i.e., self evidencing and Bayesian) accounts of brain
function. Crucially, these accounts have attendant process theories
that specify various forms of neuronal dynamics and message
passing that can be evaluated empirically
[54, 66, 70, 71, 73, 74, 176, 177]. Note that the strong kind of
computational psychiatry advocated here demands a process
theory, which can be tested empirically, in terms of the evidence
for one process theory or model, relative to another.
Although synaptopathy of a certain kind may be a common
theme—in terms of computational and pathophysiological
mechanisms—where and how these pathologies manifest is an
open question: a question that may require a different answer for
every condition (and possibly every patient). For example, at the
level of (canonical) microcircuits, neuromodulatory abnormalities
may manifest in different ways at different levels of cortical
hierarchies and, indeed, in the (inter and intralaminar) message
passing within microcircuits. This means that one needs a way of
identifying functional brain architectures—and, particularly, con-
text sensitive changes in synaptic efficacy—that report the
presence or absence of synaptopathy. In short, this calls for
in vivo and ex vivo assays of synaptic efficacy [178, 179] that can
be deployed across scale; i.e., from molecular biology through to
patients in the clinic (or bedside).
There have been amazing advances in the measurement of
neuronal processes; ranging from laminar fMRI [180, 181] through
to optogenetics [33, 91, 182–184]. From a computational
perspective, there is one universal imperative for quantifying
changes in synaptic efficacy: one has to be able to assess the
evidence—in any empirical measurement—for changes in effec-
tive connectivity. In turn, one has to commit to generative models
of how the data were generated, in order to assess the evidence
for models with and without a change [185]. Again, we come back
to the foundational role of generative models but now from the
perspective of a scientist, as opposed to the brain (as a scientist).
At its simplest, synaptic efficacy is modelled as a connection
strength (or rate) with accompanying synaptic rate or time
constants. These have to be estimated under some model of
hidden or latent physiological fluctuations that generate observed
data. This can be at a molecular scale or the scale of whole brain
imaging.
In imaging neuroscience, there are two approaches to
connectivity; namely the weak approach called functional
connectivity and the strong approach called effective connectivity
[186]. Functional connectivity is essentially looking for patterns or
correlations between measured neuronal responses. In contrast,
effective connectivity is the estimate of directed connection
strengths, under some forward or generative model. The power of
effective connectivity analyses rests on being able to compare the
Molecular Psychiatry (2023) 28:256 – 268
K. Friston
261
evidence for one model (sometimes, of functional connectivity as
a data feature) relative to another. A prescient example of this is
the estimation of intrinsic excitability or gain, usually associated
with the neuromodulatory precision control above. To assess the
intrinsic (within node or source) excitability of a node—in a
distributed network of nodes coupled by extrinsic (between node
or source) connections—one has to parameterise network models
in terms of intrinsic or self-connections [187]. This foregrounds
one of the weaknesses of descriptive or functional connectivity: in
the sense that the correlation of a source with itself tells you
nothing (it is always one).
In imaging neuroscience, the predominant analysis of effective
connectivity is dynamic causal modelling, usually based upon
neural mass models with a greater or lesser degree of biological
plausibility [16, 17, 19, 20, 22, 36, 37, 188–190]. In brief, modelling
neuronal dynamics in terms of neural mass models means that
one can parameterise intrinsic and extrinsic connectivity—and
context sensitive changes—in terms of model parameters and,
crucially, assess the evidence for models with and without certain
parameters, by fitting the models to empirical timeseries.
MEASURING SYNAPTIC EFFICACY IN VIVO
Pursuing the narrative of testing hypotheses about neuromodu-
latory synaptopathies, the three-way link between precision
control, attention (and sensory attenuation) and neuromodulation
means there has been considerable progress in characterising the
synaptic basis of sentience using attentional and oddball
paradigms. Indeed, in schizophrenia research, one of the best
studied paradigms is the mismatch negativity that allows one to
estimate excitability (i.e., intrinsic connectivity) induced by a
changing context, signalled by an oddball stimulus; e.g., [22, 191].
Clearly, changes in synaptic efficacy over a hundred milliseconds
to seconds requires electrophysiological measurements of the
kind afforded by electrophysiology (e.g., EEG and MEG). This
allows one to pinpoint the expression of aberrant gain or precision
control to various neuronal populations in cortical hierarchies. See
Fig. 4 for an early example. The dénouement of this kind of
computational study of schizophrenia (at the time of writing)
concludes: (i) EEG responses in three classic paradigms are all
attributable to the same underlying synaptic change: greater self-
inhibition in pyramidal cells; (ii) psychotic symptoms relate to
disinhibition in neural circuits and “these findings are more
commensurate with the hypothesis that a primary loss of synaptic
gain on pyramidal cells is then compensated by interneuron
downregulation (rather than the converse)” [192].
The construct validation of these synaptic assays speaks to
integration across scales. Perhaps the most developed use cases
here are in epilepsy research and the characterisation of
neurodegenerative disorders [179, 193–196]. A compelling exam-
ple of applying generative (i.e., dynamic causal) modelling across
scales can be found when drilling down on the molecular basis of
childhood epilepsy, using data from small animal models and
patients. For example, local field potential recordings in a mouse
model were used to validate a dynamic causal model of NMDAR-
Ab effects on cortical microcircuitry. The ensuing DCM was then
used to identify the synaptic parameters that best explain EEG
paroxysms in paediatric patients with NMDAR-Ab encephalitis. The
authors then returned to the mouse model to show that NMDAR-
Ab-related changes render microcircuitry critically susceptible to
overt EEG paroxysms [197]. See Fig. 5.
Another field that is pursuing a similar approach is the
characterisation of synaptopathies in neurodegeneration; again,
with convergence on the interactions between pyramidal cells
and inhibitory interneurons. For example, [198] used the mismatch
negativity paradigm to validate a dynamic causal model of
laminar-specific microcircuitry and GABAergic neurotransmission.
Bayesian model comparison identified an effect of tiagabine on
 K. Friston
262

A failure of precision control in schizophrenia:

control subjects - predictable

control subjects - unpredictable

schizophrenia - predictable
schizophrenia - unpredictable

Effects of predictability on intrinsic gain control

1.5
control subjects
log modulation 1 schizophrenics
0.5

-0.5

-1

-1.5

-2
V1 R V5 L V5 R IT L IT R PC L PC
cortical source

Fig. 4 Synaptic gain control in schizophrenia. This figure summarizes an early example of using DCM of event related responses (ERPs), to
assess the modulation of synaptic efficacy in schizophrenia using EEG [191]. ERPs were elicited in a visual target detection paradigm, under
predictable and unpredictable conditions, in neurotypical and schizophrenic subjects, respectively. Sources—modeled as small cortical
patches (lower left panel)—include: a midline visual source (V1), right and left sources near the temporoparietal junction (V5), right and left
inferotemporal sources (IT) and bilateral superior parietal sources (PC). The distributed network connecting these sources (upper right panel)
entails top-down connections from PC and IT to V5 that send backward connections to V1 (black lines); reciprocal forward connections (red
lines); and intrinsic connections for each source (black loops). The principal components of predicted and observed ERPs (upper right panel)
show a pronounced difference in the evoked responses of normal subjects to predictable and unpredictable targets around 300 to 400 ms
after stimulus onset (compare the red and blue traces). This difference is attenuated in the schizophrenia patients (green and magenta). The
lower right panel reports (log scaling of ) intrinsic connections and their 90% posterior confidence intervals, for predictable versus
unpredictable targets, for patients (white) and controls (teal). The notable thing here is a failure of predictability to modulate the intrinsic gain
(i.e., self-connectivity) throughout the hierarchy in the schizophrenic subjects.

GABAergic modulation of deep pyramidal populations and
inhibitory interneurons. The authors then applied this model to
patients with frontotemporal lobar degeneration, showing that
“the phasic inhibition of deep cortico-cortical pyramidal neurons
following tiagabine, but not placebo, was a function of GABA
concentration” [199].
Similar approaches have been considered in Parkinson’s disease
and schizophrenia, spanning cell cultures to human subjects. The
basic idea behind these proposals is to link scales through
Bayesian model comparison. In other words, optimise the model
of neuronal microcircuits in ex vivo cell cultures, in terms of
synaptic time constants of intrinsic connections, using Bayesian
model inversion and selection. One can then use the posterior
estimates as priors for the next scale (e.g., small animal
preparations, through to humans). Crucially, the superordinate
scale introduces additional parameters (e.g., extrinsic connections
among cortical areas) that are estimated more efficiently, having
resolved uncertainty about the parameters that are shared with
lower scales. In short, by recursively using yesterday’s (one scale)
posteriors as tomorrow’s (next scale) priors, one has a seamless
and principled approach to integrating across scales.
COMPUTATIONAL PHENOTYPING AND NOSOLOGY
The narrative so far has moved from the theoretical frameworks,
within which to place synaptopathy, to the requisite modelling of
empirical data to identify the synaptopathy that characterises
different disorders. This could be regarded as one kind of
precision medicine [200]; however, there is another kind that
speaks to characterising each individual in the sense of
personalised medicine. Computational approaches in this setting
are potentially important in stratifying and classifying various
clinical phenotypes for the mechanistic studies outlined above.
This usually involves summarising a subject’s behaviour in terms
of a computational model of that behaviour. Traditionally, this has
been a descriptive (e.g., reinforcement learning) model. This
represents a weak kind of phenotyping because there are no well-
developed process theories for reinforcement learning. A strong
complement to computational phenotyping can however be
formalised using something called the complete class theorem
[201, 202]. This basically says that there for any pair of behaviours
and reward functions, there exists some priors that render the
behaviour Bayes optimal. This could be read as licensing any
Bayesian ‘just so’ stories about choice behaviour [203]. However, it
has a deeper and more pragmatic implication.
It means that any given patient can be fully characterised by her
prior beliefs under ideal (active Bayesian inference) observer
assumptions. This motivates the difficult game of inferring what
generative model this subject is using, based purely upon their
behavioural or neuronal responses [204–206]. Although at an early
stage, this sort of computational phenotyping shows promise, in
the sense that it can be more efficient than simply trying to
classify or stratify patients on the basis of their responses per se
[20]. This approach, sometimes called generative embedding,

Molecular Psychiatry (2023) 28:256 – 268

 K. Friston
263

Fig. 5 Measuring synaptopathy in vivo. This figure summarises the results of a DCM study of seizure activity in mice—induced with PTZ—in
NMDA receptor antibody (Ab) negative and positive cohorts, respectively [197]. Its agenda was to examine the effects of PTZ and receptor
antibodies on synaptic efficacy; summarised here in terms of the principal [eigen] component of changes in intrinsic connectivity. The
principal component of time constants implicated superficial pyramidal and spiny stellate cell changes (A), while the principal component of
connectivity-strength reflects changes in the coupling of spiny stellate to superficial pyramidal cells (B). The two principal components (of
variations in time constants and intrinsic connectivity) constitute a parameter space summarising the modulation of synaptic efficacy. For
each point in this parameter space, one can simulate the spectral responses one would observe data space. C in this panel, estimates of
synaptic efficacy for experimental subjects in the four conditions (pre-and post-PTZ, with and without antibodies) are shown as coloured dots.
These are superimposed on the predicted log mean delta power (with selected centile isoclines). The basic message here is that seizure
induction with PTZ shifts synaptic efficacy into regimes of high delta (c.f., slow-wave activity), and that this effect is more marked in the
presence of NMDA receptor antibodies. dp, deep pyramidal cells; ii, inhibitory interneurons; sp, superficial pyramidal cells; ss, spiny stellate
cells. Adapted with permission from the authors from.

has been applied both to the generative models a subject might
use to specify her behaviour and the generative models the
researcher uses to explain her physiological or neuronal responses
[4, 207].
A related application of generative models to phenotyping
involves constructing dynamic causal models of hidden variables
or states that underwrite the signs and symptoms expressed by
patients over weeks and years. This was a notable outcome of an
Ernst Strüngmann forum on computational psychiatry [208],
which concluded that conventional diagnoses—based upon
traditional nosology—could play an important role in this kind
of computational nosology. However, the role was not to stratify
patients but rather as data features that supply evidence for or
against models of the mapping between pathophysiology and
psychopathology that are hidden from direct observation.
A key tenet of this approach is to cast pathophysiology and
psychopathology as slowly fluctuating dynamical processes. This
speaks to a separation of temporal scales between processes like
active inference [206], active learning [32] and structure learning
[209]. The mandatory coupling among these (fast and slow)
processes may underwrite cyclothymic disorders, relapsing-
remitting presentations and neurodevelopmental determinants.
An illustrative example of modelling the dynamics of schizoaffec-
tive disorders can be found in [208].
An important aspect—of this application of computational
psychiatry—is the quantification of uncertainty about how a
patient will behave, and what is likely to happen to her in the
future. Indeed, uncertainty quantification is a primary focus of
related approaches in quantitative epidemiology and meteorol-
ogy; especially in the context of forecasting and scenario
modelling [210]. In principle, finding the right generative model
for a patient guarantees the best predictions. This is because a
model with the greatest evidence precludes overfitting and
ensures generalisability [211]; namely, the generalisation from old
(i.e., what has happened to the patient in the past) to new data
(what will happen to her in the future).
COMPUTATIONAL NEUROPSYCHOLOGY
A final part of the narrative is that a generative model of
pathophysiology, and ensuing psychopathology, allows one to
perform in silico or synthetic experiments [212–214]. These entail
optimising the parameters of a model of a particular subject or
cohort, and seeing what would happen if one increased or
decreased certain synaptic efficacies, e.g., [197]. Alternatively, one
can emulate extremes of aberrant precision by simply deleting
connections in active inference models of diagnostic paradigms,
e.g., [215]. In these computational studies, the extrinsic (between
node) connections are usually assigned to likelihood mappings
relating sensory observations to representations or expectations
about hidden states of affairs. Conversely, intrinsic (within node)
connections are usually treated as embodying prior beliefs about
state transitions and narratives that characterise the paradigm in
question.
One obvious advantage of being able to create a ‘digital twin’ of
a patient in silico, is the ability to perform synthetic lesion studies
and psychopharmacology, to test various hypotheses about the
effect of therapeutic interventions. This is probably best illustrated
in the setting of epilepsy, particularly, in predicting the effects of
pharmacological and surgical interventions; e.g., [216].

Molecular Psychiatry (2023) 28:256 – 268

 K. Friston
264
CONCLUSION
This review has taken a somewhat colloquial tour through the
fundaments of one (strong) kind of computational psychiatry.
Perhaps the take-home message is the foundational importance
of a hypothesis or generative model—and being able to articulate
this model in computational or formal terms. This is a difficult
problem; both in terms of identifying or selecting the right sort of
generative models that explain sentient behaviour and its
pathologies, but also the biophysical (e.g., dynamic causal) models
we, as researchers, use to make sense of our data. This difficulty
should not be understated. For example, although dynamic causal
modelling has enjoyed wide uptake in fMRI, the requisite models
—necessary to explain fine-grained neuronal dynamics in
electrophysiology—are notoriously difficult to build and explore
using model comparison (a.k.a., structure learning). This usually
calls for researchers that are well versed in the system and scale of
enquiry, who are also fluent in Bayesian and variational modelling
procedures.
The ultimate goal of this endeavour is a generative model or
explanation for various psychiatric disorders that furnishes a
parsimonious yet accurate account of all the data at hand. This
truism reflects the fact that model evidence can be decomposed
into accuracy minus complexity; speaking to the pressure to find
minimally complex explanations that are as simple as possible but
not too simple. In short, the mechanistic explanations we seek will
be alluringly simple, but the journey may be difficult. One could
argue that psychiatry has chosen the most difficult journey of all;
namely, to explain ourselves.
REFERENCES
1. Huys QJ, Moutoussis M, Williams J. Are computational models of any use to
psychiatry? Neural Netw. 2011;24:544–51.
2. Montague PR, Dolan RJ, Friston KJ, Dayan P. Computational psychiatry. Trends
Cogn Sci. 2012;16:72–80.
3. Corlett PR, Fletcher PC. Computational psychiatry: a Rosetta Stone linking the
brain to mental illness. Lancet Psychiatry. 2014;1:399–402.
4. Stephan KE, Mathys C. Computational approaches to psychiatry. Curr Opin
Neurobiol. 2014;25:85–92.
5. Wang XJ, Krystal JH. Computational psychiatry. Neuron. 2014;84:638–54.
6. Liddle P, Carpenter WT, Crow T. Syndromes of Schizophrenia: Classic Literature.
Br J Psychiatry. 1994;165:721–7.
7. Friston KJ, Liddle PF, Frith CD, Hirsch SR, Frackowiak RS. The left medial temporal
region and schizophrenia. A PET study. Brain. 1992;115:367–82.
8. Sutton RS, Barto AG. Reinforcement Learning: An Introduction. MIT Press: Cam-
bridge, MA, 1998.
9. MacKay DJC. Information-based objective functions for active data selection.
Neural Comput. 1992;4:590–604.
10. Active Motor Babbling for Sensorimotor Learning. Proceedings of the 2008 Ieee
International Conference on Robotics and Biomimetics, Vols 1–4. 2009.
11. Schmidhuber J. Formal Theory of Creativity, Fun, and Intrinsic Motivation
(1990–2010). Ieee Trans Autonomous Ment Dev. 2010;2:230–47.
12. Barto A, Mirolli M, Baldassarre G. Novelty or surprise? Front Psychol. 2013;4:907.
13. Oudeyer P-Y, Kaplan F. What is intrinsic motivation? a typology of computa-
tional approaches. Front Neurorobotics. 2007;1:6.
14. Schwartenbeck P, Passecker J, Hauser TU, FitzGerald TH, Kronbichler M, Friston
KJ. Computational mechanisms of curiosity and goal-directed exploration. eLife.
2019;8:e41703.
15. Friston KJ, Frith CD, Liddle PF, Frackowiak RS. Functional connectivity: the
principal-component analysis of large (PET) data sets. J Cereb Blood Flow Metab.
1993;13:5–14.
16. Benetti S, Mechelli A, Picchioni M, Broome M, Williams S, McGuire P. Functional
integration between the posterior hippocampus and prefrontal cortex is
impaired in both first episode schizophrenia and the at risk mental state. Brain.
2009;132:2426–36.
17. Allen P, Stephan KE, Mechelli A, Day F, Ward N, Dalton J, et al. Cingulate activity
and fronto-temporal connectivity in people with prodromal signs of psychosis.
Neuroimage. 2010;49:947–55.
18. Tan HY, Chen AG, Kolachana B, Apud JA, Mattay VS, Callicott JH, et al.
Effective connectivity of AKT1-mediated dopaminergic working memory net-
works and pharmacogenetics of anti-dopaminergic treatment. Brain.
2012;135:1436–45.
19. Dauvermann MR, Whalley HC, Romaniuk L, Valton V, Owens DG, Johnstone EC,
et al. The application of nonlinear Dynamic Causal Modelling for fMRI in subjects
at high genetic risk of schizophrenia. Neuroimage. 2013;73:16–29.
20. Brodersen KH, Deserno L, Schlagenhauf F, Lin Z, Penny WD, Buhmann JM, et al.
Dissecting psychiatric spectrum disorders by generative embedding. Neuro-
Image Clin. 2014;4:98–111.
21. Kaplan CM, Saha D, Molina JL, Hockeimer WD, Postell EM, Apud JA, et al. Esti-
mating changing contexts in schizophrenia. Brain. 2016;139:2082–95.
22. Ranlund S, Adams RA, Diez A, Constante M, Dutt A, Hall MH, et al. Impaired
prefrontal synaptic gain in people with psychosis and their relatives during the
mismatch negativity. Hum Brain Mapp. 2016;37:351–65.
23. Cortes-Briones JA, Tapia-Rivas NI, D’Souza DC, Estevez PA. Going deep into
schizophrenia with artificial intelligence. Schizophrenia Res. 2022;245:122–40
24. Marcinkevičs R, Vogt JE. Interpretability and explainability: a machine learning
Zoo Mini-tour. 2020, p arXiv:2012.01805.
25. Bai J, Song Q, Cheng G. Efficient variational inference for sparse deep learning
with theoretical guarantee. 2020, p arXiv:2011.07439.
26. Millidge B. Deep active inference as Variational Policy Gradients. arXiv e-prints
2019: arXiv:1907.03876.
27. Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative
Adversarial Networks. Proceedings of the 34th International Conference on
Machine Learning; Aug 06-11 2017; Sydney, AUSTRALIA2017.
28. Planning by Probabilistic Inference. Proceedings of the Proc. of the 9th Int.
Workshop on Artificial Intelligence and Statistics 2003.
29. Botvinick M, Toussaint M. Planning as inference. Trends Cogn Sci.
2012;16:485–8.
30. Da Costa L, Parr T, Sajid N, Veselic S, Neacsu V, Friston K. Active inference on
discrete state-spaces: a synthesis. J Math Psychol. 2020;99:102447.
31. Lanillos P, Meo C, Pezzato C, Meera AA, Baioumy M, Ohata W, et al. Active
Inference in Robotics and Artificial Agents: Survey and Challenges. 2021, p
arXiv:2112.01871.
32. MacKay DJC Information Theory, Inference and Learning Algorithms. Cambridge
University Press: Cambridge, 2003.
33. Obi-Nagata K, Temma Y, Hayashi-Takagi A. Synaptic functions and their dis-
ruption in schizophrenia: From clinical evidence to synaptic optogenetics in an
animal model. Proc Jpn Acad Ser B Phys Biol Sci. 2019;95:179–97.
34. Stephan KE, Friston KJ, Frith CD. Dysconnection in schizophrenia: from abnormal
synaptic plasticity to failures of self-monitoring. Schizophr Bull. 2009;35:509–27.
35. Friston K, Brown HR, Siemerkus J, Stephan KE. The dysconnection hypothesis
(2016). Schizophr Res. 2016;176:83–94.
36. Dima D, Frangou S, Burge L, Braeutigam S, James AC. Abnormal intrinsic and
extrinsic connectivity within the magnetic mismatch negativity brain network in
schizophrenia: a preliminary study. Schizophr Res. 2012;135:23–7.
37. Schmidt A, Smieskova R, Simon A, Allen P, Fusar-Poli P, McGuire PK, et al.
Abnormal effective connectivity and psychopathological symptoms in the
psychosis high-risk state. J Psychiatry Neurosci: JPN. 2014;39:239–48.
38. Helmholtz H. Concerning the perceptions in general. Treatise on physiological
optics, vol. III. Dover: New York, 1866/1962.
39. Helmholtz H. The Facts of Perception. In: Middletown RK (ed). The Selected
Writings of Hermann von Helmholtz. Wesleyan University Press: Connecticut,
1878, p 384.
40. Yuille A, Kersten D. Vision as Bayesian inference: analysis by synthesis? Trends
Cogn Sci. 2006;10:301–8.
41. Neisser U. Cognitive psychology. Appleton-Century-Crofts: New York, 1967.
42. MacKay DMC. The epistemological problem for automata. In: Shannon C,
McCarthy J (eds). Automata Studies. Princeton University Press: Princeton, NJ,
1956, pp 235–51.
43. Gregory RL. Perceptual illusions and brain models. Proc R Soc Lond B.
1968;171:179–96.
44. Gregory RL. Perceptions as hypotheses. Philos Trans R Soc Lond B.
1980;290:181–97.
45. Dayan P, Hinton GE, Neal RM, Zemel RS. The Helmholtz machine. Neural
Comput. 1995;7:889–904.
46. Hutto D, Myin E. Radicalizing enactivism: Basic minds without content. MIT
Press: Cambridge, MA, 2013.
47. Clark A, Chalmers DJ. The extended mind. Analysis. 1998;58:7–19.
48. Clark A. Being there putting brain, body, and world together again. MIT Press:
Cambridge, Mass. 2001.
49. Goodwin C. Action and embodiment within situated human interaction. J
Pragmat. 2000;32:1489–522.
50. Barsalou LW. Grounded cognition. Annu Rev Psychol. 2008;59:617–45.
51. Chemero A. Radical Embodied Cognitive Science, 2009, pp 1–252.
52. Seth AK. A predictive processing theory of sensorimotor contingencies:
Explaining the puzzle of perceptual presence and its absence in synesthesia.
Cogn Neurosci. 2014;5:97–118.
Molecular Psychiatry (2023) 28:256 – 268

53. Wiese W. Action Is Enabled by Systematic Misrepresentations. Erkenntnis.
2017;82:1233–52.
54. Parr T, Friston KJ. The Anatomy of Inference: Generative Models and Brain
Structure. Front Comput Neurosci. 2018;12:90
55. Hohwy J. New directions in predictive processing. Mind Language.
2020;35:209–33
56. Rao RP, Ballard DH. Predictive coding in the visual cortex: a functional inter-
pretation of some extra-classical receptive-field effects. Nat Neurosci.
1999;2:79–87.
57. Ballard DH, Kit D, Rothkopf CA, Sullivan B. A Hierarchical Modular Architecture
for Embodied Cognition. Multisens Res. 2013;26:177.
58. Friston KJ, Daunizeau J, Kilner J, Kiebel SJ. Action and behavior: a free-energy
formulation. Biol Cybern. 2010;102:227–60.
59. Clark A. Whatever next? Predictive brains, situated agents, and the future of
cognitive science. Behav Brain Sci. 2013;36:181–204.
60. Hohwy J. The Predictive Mind. Oxford University Press: Oxford, 2013.
61. Knill DC, Pouget A. The Bayesian brain: the role of uncertainty in neural coding
and computation. Trends Neurosci. 2004;27:712–9.
62. Doya K. Bayesian brain: Probabilistic approaches to neural coding. MIT press
2007.
63. Hohwy J. The Self-Evidencing. Brain Noûs. 2016;50:259–85.
64. Elias P. Predictive coding–I. IRE Trans Inf Theory. 1955;1:16–24.
65. Srinivasan MV, Laughlin SB, Dubs A. Predictive coding: a fresh view of inhibition
in the retina. Proc R Soc Lond B Biol Sci. 1982;216:427–59.
66. Bastos AM, Usrey WM, Adams RA, Mangun GR, Fries P, Friston KJ. Canonical
microcircuits for predictive coding. Neuron. 2012;76:695–711.
67. Egner T, Summerfield C. Grounding predictive coding models in empirical
neuroscience research. Behav Brain Sci. 2013;36:210–1.
68. Winn J, Bishop CM. Variational message passing. J Mach Learn Res.
2005;6:661–94.
69. On Variational Message Passing on Factor Graphs. Proceedings of the 2007 IEEE
International Symposium on Information Theory; 24–29 June 2007, 2007.
70. Friston K, Parr T, de Vries B. The graphical brain: Belief propagation and active
inference. Netw Neurosci (Camb, Mass). 2017;1:381–414.
71. Parr T, Markovic D, Kiebel SJ, Friston KJ. Neuronal message passing using Mean-
field, Bethe, and Marginal approximations. Sci Rep. 2019;9:1889.
72. Da Costa L, Parr T, Sengupta B, Friston K. Neural Dynamics under Active Infer-
ence: Plausibility and Efficiency of Information Processing. Entropy (Basel).
2021;23:454
73. Adams RA, Shipp S, Friston KJ. Predictions not commands: active inference in
the motor system. Brain Struct Funct. 2013;218:611–43.
74. Shipp S. Neural Elements for Predictive Coding. Front Psychol. 2016;7:1792.
75. Mumford D. On the computational architecture of the neocortex. II. Biol Cyber.
1992;66:241–51.
76. Lee TS, Mumford D. Hierarchical Bayesian inference in the visual cortex. J Optical
Soc Am A, Opt, image Sci, Vis. 2003;20:1434–48.
77. Markov N, Ercsey-Ravasz M, Van Essen D, Knoblauch K, Toroczkai Z, Kennedy H.
Cortical high-density counterstream architectures. Science. 2013;342:1238406.
78. Pinotsis DA, Brunet N, Bastos A, Bosman CA, Litvak V, Fries P, et al. Contrast gain
control and horizontal interactions in V1: a DCM study. Neuroimage.
2014;92:143–55.
79. Shaw AD, Moran RJ, Muthukumaraswamy SD, Brealy J, Linden DE, Friston KJ,
et al. Neurophysiologically-informed markers of individual variability and
pharmacological manipulation of human cortical gamma. Neuroimage.
2017;161:19–31.
80. Kok P, Rahnev D, Jehee JFM, Lau HC, de Lange FP. Attention Reverses the Effect
of Prediction in Silencing Sensory Signals. Cereb Cortex. 2012;22:2197–206.
81. Auksztulewicz R, Friston K. Attentional Enhancement of Auditory Mismatch
Responses: a DCM/MEG Study. Cereb Cortex. 2015;25:4273–83.
82. Ainley V, Apps MA, Fotopoulou A, Tsakiris M. ‘Bodily precision’: a predictive
coding account of individual differences in interoceptive accuracy. Philosophical
transactions of the Royal Society of London Series B, Biological sciences.
2016;371:20160003
83. Limanowski J. Precision control for a flexible body representation. Neurosci
Biobehavioral Rev. 2021;134:104401.
84. Clark A. The many faces of precision (Replies to commentaries on “Whatever
next? Neural prediction, situated agents, and the future of cognitive science”).
Front Psychol. 2013;4:270.
85. Feldman H, Friston KJ. Attention, uncertainty, and free-energy. Front Hum
Neurosci. 2010;4:215.
86. Graboi D, Lisman J. Recognition by top-down and bottom-up processing in
cortex: the control of selective attention. J Neurophysiol. 2003;90:798–810.
87. Lisman J, Buzsaki G. A neural coding scheme formed by the combined function
of gamma and theta oscillations. Schizophr Bull. 2008;34:974–80.
Molecular Psychiatry (2023) 28:256 – 268
K. Friston
265
88. Lisman J. Excitation, inhibition, local oscillations, or large-scale loops: what
causes the symptoms of schizophrenia? Curr Opin Neurobiol. 2012;22:537–44.
89. Bauer M, Stenner MP, Friston KJ, Dolan RJ. Attentional modulation of alpha/beta
and gamma oscillations reflect functionally distinct processes. J Neurosci.
2014;34:16117–25.
90. Spencer KM, Nestor PG, Niznikiewicz MA, Salisbury DF, Shenton ME, McCarley
RW. Abnormal neural synchrony in schizophrenia. J Neurosci. 2003;23:7407–11.
91. Sohal VS, Zhang F, Yizhar O, Deisseroth K. Parvalbumin neurons and gamma
rhythms enhance cortical circuit performance. Nature. 2009;459:698–702.
92. Parr T, Friston KJ. Uncertainty, epistemics and active inference. J R Soc Interface.
2017;14:20170376.
93. Brown H, Adams RA, Parees I, Edwards M, Friston K. Active inference, sensory
attenuation and illusions. Cogn Process. 2013;14:411–27.
94. Oestreich LK, Mifsud NG, Ford JM, Roach BJ, Mathalon DH, Whitford TJ. Sub-
normal sensory attenuation to self-generated speech in schizotypy: Electro-
physiological evidence for a ‘continuum of psychosis’. Int J Psychophysiol: Off J
Int Organ Psychophysiol. 2015;97:131–8.
95. Limanowski J (Dis-)Attending to the Body. In: Metzinger TK, Wiese W (eds).
Philosophy and Predictive Processing. MIND Group: Frankfurt am Main, 2017.
96. Blakemore SJ, Frith CD, Wolpert DM. Spatio-temporal prediction modulates the
perception of self-produced stimuli. J Cogn Neurosci. 1999;11:551–9.
97. Shergill SS, Samson G, Bays PM, Frith CD, Wolpert DM. Evidence for sensory
prediction deficits in schizophrenia. Am J Psychiatry. 2005;162:2384–6.
98. Hughes G, Desantis A, Waszak F. Mechanisms of intentional binding and sensory
attenuation: the role of temporal prediction, temporal control, identity predic-
tion, and motor prediction. Psychol Bull. 2013;139:133–51.
99. Wurtz RH. Neuronal mechanisms of visual stability. Vis Res. 2008;48:2070–89.
100. Adams RA, Stephan KE, Brown HR, Frith CD, Friston KJ. The computational
anatomy of psychosis. Front Psychiatry. 2013;4:47.
101. Fuster JM. Upper processing stages of the perception-action cycle. Trends Cogn
Sci. 2004;8:143–5.
102. Wurtz RH, McAlonan K, Cavanaugh J, Berman RA. Thalamic pathways for active
vision. Trends Cogn Sci. 2011;5:177–84.
103. Pillmann F. Carl Wernicke and the neurobiological paradigm in psychiatry. Acta
Neuropsychologica. 2007;5:246–60.
104. Catani M, Ffytche DH. The rises and falls of disconnection syndromes. Brain.
2005;128:2224–39.
105. Bob P, Mashour GA. Schizophrenia, dissociation, and consciousness. Conscious
Cogn. 2011;20:1042–9.
106. Weinberger DR. A connectionist approach to the prefrontal cortex. J Neu-
ropsychiatry Clin Neurosci. 1993;5:241–53.
107. Friston KJ, Frith CD. Schizophrenia: a disconnection syndrome? Clin Neurosci.
1995;3:89–97.
108. Grant SG. Synaptopathies: diseases of the synaptome. Curr Opin Neurobiol.
2012;22:522–9.
109. Nicastro N, Holland N, Savulich G, Carter SF, Mak E, Hong YT, et al. (11)C-UCB-J
synaptic PET and multimodal imaging in dementia with Lewy bodies. Eur J
Hybrid Imaging. 2020;4:25–5.
110. Cohen JD, Servan-Schreiber D. Context, cortex, and dopamine: a connectionist
approach to behavior and biology in schizophrenia. Psychol Rev. 1992;99:45–77.
111. Winterer G, Weinberger DR. Genes, dopamine and cortical signal-to-noise ratio
in schizophrenia. Trends Neurosci. 2004;27:683–90.
112. Laruelle M. Schizophrenia: from dopaminergic to glutamatergic interventions.
Curr Opin Pharm. 2014;14C:97–102.
113. Geyer MA, Krebs-Thomson K, Braff DL, Swerdlow NR. Pharmacological studies of
prepulse inhibition models of sensorimotor gating deficits in schizophrenia: a
decade in review. Psychopharmacology. 2001;156:117–54.
114. Krystal JH, Karper LP, Seibyl JP, Freeman GK, Delaney R, Bremner JD, et al.
Subanesthetic effects of the noncompetitive NMDA antagonist, ketamine, in
humans. Psychotomimetic, perceptual, cognitive, and neuroendocrine respon-
ses. Arch Gen Psychiatry. 1994;51:199–214.
115. Olney JW, Newcomer JW, Farber NB. NMDA receptor hypofunction model of
schizophrenia. J Psychiatr Res. 1999;33:523–33.
116. Gonzalez-Burgos G, Lewis DA. NMDA receptor hypofunction, parvalbumin-
positive neurons, and cortical gamma oscillations in schizophrenia. Schizophr
Bull. 2012;38:950–7.
117. Lisman JE, Coyle JT, Green RW, Javitt DC, Benes FM, Heckers S, et al. Circuit-
based framework for understanding neurotransmitter and risk gene interactions
in schizophrenia. Trends Neurosci. 2008;31:234–42.
118. Uhlhaas PJ, Singer W. Abnormal neural oscillations and synchrony in schizo-
phrenia. Nat Rev Neurosci. 2010;11:100–13.
119. Jardri R, Hugdahl K, Hughes M, Brunelin J, Waters F, Alderson-Day B, et al. Are
Hallucinations Due to an Imbalance Between Excitatory and Inhibitory Influ-
ences on the Brain? Schizophrenia Bull. 2016;42:1124–34.
 K. Friston
266
120. Schizophrenia Working Group of the Psychiatric Genomics C. Biological insights
from 108 schizophrenia-associated genetic loci. Nature. 2014;511:421–7.
121. Hall J, Bray NJ. Schizophrenia Genomics: Convergence on Synaptic Develop-
ment, Adult Synaptic Plasticity, or Both? Biol Psychiatry. 2022;91:709–17
122. Trubetskoy V, Pardiñas AF, Qi T, Panagiotaropoulou G, Awasthi S, Bigdeli TB,
et al. Mapping genomic loci implicates genes and synaptic biology in schizo-
phrenia. Nature. 2022;604:502–8.
123. Singh T, Poterba T, Curtis D, Akil H, Al Eissa M, Barchas JD, et al. Rare coding
variants in ten genes confer substantial risk for schizophrenia. Nature.
2022;604:509–16.
124. Sanacora G, Treccani G, Popoli M. Towards a glutamate hypothesis of depres-
sion An emerging frontier of neuropsychopharmacology for mood disorders.
Neuropharmacology. 2012;62:63–77.
125. Hamon M, Blier P. Monoamine neurocircuitry in depression and strategies for
new treatments. Prog Neuropsychopharmacol Biol Psychiatry. 2013;45:54–63.
126. Russo M, Carrarini C, Dono F, Rispoli MG, Di Pietro M, Di Stefano V, et al. The
pharmacology of visual hallucinations in synucleinopathies. Front Pharm.
2019;10:1379.
127. Collerton D, Perry E, McKeith I. Why people see things that are not there: a novel
Perception and Attention Deficit model for recurrent complex visual halluci-
nations. Behav Brain Sci. 2005;28:737–57.
128. Friston KJ. Hallucinations and perceptual inference. Behav Brain Sci.
2005;28:764–6.
129. Fletcher PC, Frith CD. Perceiving is believing: a Bayesian approach to explaining
the positive symptoms of schizophrenia. Nat Rev Neurosci. 2009;10:48–58.
130. Teufel C, Fletcher P, Davis G. Seeing other minds: attributed mental states
influence perception. Trends Cogn Sci. 2010;14:376–82.
131. Corlett PR, Honey GD, Krystal JH, Fletcher PC. Glutamatergic model psychoses:
prediction error, learning, and inference. Neuropsychopharmacology.
2011;36:294–315.
132. Haarsma J, Fletcher PC, Ziauddeen H, Spencer TJ, Diederen KMJ, Murray GK.
Precision weighting of cortical unsigned prediction errors is mediated by
dopamine and benefits learning. bioRxiv. 2018.
133. Sterzer P, Adams RA, Fletcher P, Frith C, Lawrie SM, Muckli L, et al. The predictive
coding account of psychosis. Biol Psychiatry. 2018;84:634–43.
134. Pellicano E, Burr D. When the world becomes ‘too real’: a Bayesian explanation
of autistic perception. Trends Cogn Sci. 2012;16:504–10.
135. Lawson RP, Rees G, Friston KJ. An aberrant precision account of autism. Front
Hum Neurosci. 2014;8:302.
136. Van de Cruys S, Evers K, Van der Hallen R, Van Eylen L, Boets B, de-Wit L, et al.
Precise minds in uncertain worlds: predictive coding in autism. Psychol Rev.
2014;121:649–75.
137. Edwards MJ, Adams RA, Brown H, Parees I, Friston KJ. A Bayesian account of
‘hysteria’. Brain. 2012;135:3495–512.
138. Skewes JC, Jegindo EM, Gebauer L. Perceptual inference and autistic traits.
Autism. 2014: [Epub ahead of print].
139. Gu X, Eilam-Stock T, Zhou T, Anagnostou E, Kolevzon A, Soorya L, et al. Auto-
nomic and brain responses associated with empathy deficits in autism spectrum
disorder. Hum Brain Mapp. 2015;36:3323–38.
140. Clark JE, Watson S, Friston KJ. What is mood? A computational perspective.
Psychol Med. 2018;48:2277–84.
141. Benrimoh D, Parr T, Adams RA, Friston K. Hallucinations both in and out of
context: an active inference account. PLoS One. 2019;14:e0212379.
142. Krahe C, Springer A, Weinman JA, Fotopoulou A. The social modulation of pain:
others as predictive signals of salience - a systematic review. Front Hum Neu-
rosci. 2013;7:386.
143. Hoskin R, Berzuini C, Acosta-Kane D, El-Deredy W, Guo H, Talmi D. Sensitivity to
pain expectations: A Bayesian model of individual differences. Cognition.
2018;182:127–39.
144. Gerrans P. Pain Asymbolia as depersonalization for pain experience. An Inter-
oceptive Active Inference Account. Front Psychol. 2020;11:523710
145. Seymour B, Mancini F. Hierarchical models of pain: Inference, information-
seeking, and adaptive control. NeuroImage. 2020;222:117212.
146. Paulus MP, Stein MB. An insular view of anxiety. Biol Psychiatry. 2006;60:383–7.
147. Cornwell BR, Garrido MI, Overstreet C, Pine DS, Grillon C. The Unpredictive Brain
Under Threat: A Neurocomputational Account of Anxious Hypervigilance. Biol
Psychiatry. 2017;82:447–54.
148. Duquette P. Increasing Our Insular World View: Interoception and Psycho-
pathology for Psychotherapists. Front Neurosci. 2017;11:135.
149. Smith R, Kuplicki R, Feinstein J, Forthman KL, Stewart JL, Paulus MP, et al. A
Bayesian computational model reveals a failure to adapt interoceptive precision
estimates across depression, anxiety, eating, and substance use disorders. PLOS
Computational Biol. 2020;16:e1008484.
150. Notredame CE, Pins D, Deneve S, Jardri R. What visual illusions teach us about
schizophrenia. Front Integr Neurosci. 2014;8:63.
151. Powers AR 3rd, Gancsos MG, Finn ES, Morgan PT, Corlett PR. Ketamine-Induced
Hallucinations. Psychopathology. 2015;48:376–85.
152. Teufel C, Subramaniam N, Dobler V, Perez J, Finnemann J, Mehta PR, et al.
Shift toward prior knowledge confers a perceptual advantage in early psychosis
and psychosis-prone healthy individuals. Proc Natl Acad Sci USA.
2015;112:13401–6.
153. Powers AR, Mathys C, Corlett PR. Pavlovian conditioning–induced hallucinations
result from overweighting of perceptual priors. Science. 2017;357:596–600.
154. Moutoussis M, Bentall RP, El-Deredy W, Dayan P. Bayesian modelling of
Jumping-to-Conclusions bias in delusional patients. Cogn Neuropsychiatry.
2011;16:422–47.
155. Joyce DW, Averbeck BB, Frith CD, Shergill SS. Examining belief and confidence in
schizophrenia. Psychol Med. 2013;43:2327–38.
156. Paulus MP, Feinstein JS, Khalsa SS. An Active Inference Approach to Inter-
oceptive Psychopathology. In: Widiger T, Cannon TD (eds). Annual Review of
Clinical Psychology, Vol 15, vol. 152019, pp 97–122.
157. Rae CL, Critchley HD, Seth AK. A Bayesian Account of the Sensory-Motor Inter-
actions Underlying Symptoms of Tourette Syndrome. Front Psychiatry. 2019;10:29
158. Stuke H, Weilnhammer VA, Sterzer P, Schmack K. Delusion Proneness is Linked
to a Reduced Usage of Prior Beliefs in Perceptual Decisions. Schizophrenia Bull.
2019;45:80–6.
159. Kube T, Rozenkrantz L, Rief W, Barsky A. Understanding persistent physical
symptoms: Conceptual integration of psychological expectation models and
predictive processing accounts. Clin Psychol Rev. 2020;76:101829.
160. Lyndon S, Corlett PR. Hallucinations in Posttraumatic Stress Disorder: Insights
From Predictive Coding. J Abnorm Psychol. 2020;129:534–43.
161. Richards KL, Karvelis P, Lawrie SM, Series P. Visual statistical learning and inte-
gration of perceptual priors are intact in attention deficit hyperactivity disorder.
Plos One. 2020;15:e0243100
162. Frith CD, Blakemore SJ, Wolpert DM. Abnormalities in the awareness and control
of action. Philos Trans R Soc Lond Ser B, Biol Sci. 2000;355:1771–88.
163. Shergill SS, White TP, Joyce DW, Bays PM, Wolpert DM, Frith CD. Functional
magnetic resonance imaging of impaired sensory prediction in schizophrenia.
JAMA Psychiatry. 2014;71:28–35.
164. Parees I, Brown H, Nuruki A, Adams RA, Davare M, Bhatia KP, et al. Loss of
sensory attenuation in patients with functional (psychogenic) movement dis-
orders. Brain. 2014;137:2916–21.
165. Bhat A, Parr T, Ramstead M, Friston K. Immunoceptive inference: why are psy-
chiatric disorders and immune responses intertwined? Biol Philos. 2021;36:27.
166. Quattrocki E, Friston K. Autism, oxytocin and interoception. Neurosci Biobehav
Rev. 2014;47c:410–30.
167. Palmer CJ, Paton B, Kirkovski M, Enticott PG, Hohwy J. Context sensitivity in
action decreases along the autism spectrum: a predictive processing perspec-
tive. Proc R Soc B: Biol Sci. 2015;282:20141557.
168. Palmer CJ, Seth AK, Hohwy J. The felt presence of other minds: Predictive proces-
sing, counterfactual predictions, and mentalising in autism. Conscious Cogn. 2015.
169. Lawson RP, Mathys C, Rees G. Adults with autism overestimate the volatility of
the sensory environment. Nat Neurosci. 2017;20:1293–9.
170. Palmer CJ, Lawson RP, Hohwy J. Bayesian approaches to autism: Towards
volatility, action, and behavior. Psychol Bull. 2017;143:521–42.
171. Coltheart M, Menzies P, Sutton J. Abductive inference and delusional belief.
Cogn Neuropsychiatry. 2010;15:261–87.
172. Adams RA, Vincent P, Benrimoh D, Friston KJ, Parr T. Everything is connected:
Inference and attractors in delusions. Schizophr Res. 2022;245:5–22.
173. Butler PD, Silverstein SM, Dakin SC. Visual perception and its impairment in
schizophrenia. Biol Psychiatry. 2008;64:40–7.
174. Seth AK, Friston KJ. Active interoceptive inference and the emotional brain.
Philos Trans R Soc Lon B Biol Sci. 2016;371:20160007.
175. Averbeck BB, Evans S, Chouhan V, Bristow E, Shergill SS. Probabilistic learning
and inference in schizophrenia. Schizophr Res. 2011;127:115–22.
176. Kanai R, Komura Y, Shipp S, Friston K. Cerebral hierarchies: predictive proces-
sing, precision and the pulvinar. Philosophical Trans R Soc Lond B Biol Sci.
2015;370:20140169.
177. Friston K, FitzGerald T, Rigoli F, Schwartenbeck P, Pezzulo G. Active Inference: A
Process Theory. Neural Comput. 2017;29:1–49.
178. Moran RJ, Symmonds M, Stephan KE, Friston KJ, Dolan RJ. An in vivo assay of
synaptic function mediating human cognition. Curr Biol. 2011;21:1320–5.
179. Shaw AD, Hughes LE, Moran R, Coyle-Gilchrist I, Rittman T, Rowe JB. In Vivo
Assay of Cortical Microcircuitry in Frontotemporal Dementia: A Platform for
Experimental Medicine Studies. Cereb Cortex. 2021;31:1837–47.
180. Self MW, van Kerkoerle T, Goebel R, Roelfsema PR. Benchmarking laminar fMRI:
Neuronal spiking and synaptic activity during top-down and bottom-up pro-
cessing in the different layers of cortex. Neuroimage. 2019;197:806–17.
181. Haarsma J, Kok P, Browning M. The promise of layer-specific neuroimaging for
testing predictive coding theories of psychosis. Schizophrenia Res. 2020.
Molecular Psychiatry (2023) 28:256 – 268

182. Kravitz AV, Freeze BS, Parker PR, Kay K, Thwin MT, Deisseroth K, et al. Regulation
of parkinsonian motor behaviours by optogenetic control of basal ganglia cir-
cuitry. Nature. 2010;466:622–6.
183. Anenberg E, Chan AW, Xie Y, LeDue JM, Murphy TH. Optogenetic stimulation of
GABA neurons can decrease local neuronal activity while increasing cortical
blood flow. J Cereb Blood Flow Metab: Off J Int Soc Cereb Blood Flow Metab.
2015;35:1579–86.
184. Okun M, Steinmetz NA, Cossell L, Iacaruso MF, Ko H, Bartho P, et al. Diverse
coupling of neurons to populations in sensory cortex. Nature. 2015;521:511–5.
185. Penny WD. Comparing dynamic causal models using AIC, BIC and free energy.
Neuroimage. 2012;59:319–30.
186. Friston KJ. Functional and effective connectivity: a review. Brain Connectivity.
2011;1:13–36.
187. Snyder AD, Ma L, Steinberg JL, Woisard K, Moeller FG. Dynamic Causal Modeling
Self-Connectivity Findings in the Functional Magnetic Resonance Imaging
Neuropsychiatric Literature. Front Neurosci 2021;15:636273.
188. Curcic-Blake B, van der Meer L, Pijnenborg GH, David AS, Aleman A. Insight and
psychosis: Functional and anatomical brain connectivity and self-reflection in
Schizophrenia. Hum Brain Mapp. 2015;36:4859–68.
189. Bastos-Leite AJ, Ridgway GR, Silveira C, Norton A, Reis S, Friston KJ. Dyscon-
nectivity within the default mode in first-episode schizophrenia: a stochastic
dynamic causal modeling study with functional magnetic resonance imaging.
Schizophr Bull. 2015;41:144–53.
190. Ranlund S, Nottage J, Shaikh M, Dutt A, Constante M, Walshe M, et al. Resting
EEG in psychosis and at-risk populations-a possible endophenotype? Schizophr
Res. 2014;153:96–102.
191. Fogelson N, Litvak V, Peled A, Fernandez-del-Olmo M, Friston K. The functional
anatomy of schizophrenia: A dynamic causal modeling study of predictive
coding. Schizophr Res. 2014;158:204–12.
192. Adams RA, Pinotsis D, Tsirlis K, Unruh L, Mahajan A, Horas AM, et al. Compu-
tational Modeling of Electroencephalography and Functional Magnetic Reso-
nance Imaging Paradigms Indicates a Consistent Loss of Pyramidal Cell Synaptic
Gain in Schizophrenia. Biol Psychiatry. 2022;91:202–15.
193. Rosch R, Burrows DRW, Jones LB, Peters CH, Ruben P, Samarut E. Functional
Genomics of Epilepsy and Associated Neurodevelopmental Disorders Using
Simple Animal Models: From Genes, Molecules to Brain Networks. Front Cell
Neurosci. 2019;13:556.
194. Rosch RE, Hunter PR, Baldeweg T, Friston KJ, Meyer MP. Calcium imaging and
dynamic causal modelling reveal brain-wide changes in effective connectivity
and synaptic dynamics during epileptic seizures. PLoS Comput Biol.
2018;14:e1006375.
195. Peters C, Rosch RE, Hughes E, Ruben PC. Temperature-dependent changes in
neuronal dynamics in a patient with an SCN1A mutation and hyperthermia
induced seizures. Sci Rep. 2016;6:31879.
196. Rowe JB, Hughes LE, Barker RA, Owen AM. Dynamic causal modelling of
effective connectivity from fMRI: are results reproducible and sensitive to Par-
kinson’s disease and its treatment? Neuroimage. 2010;52:1015–26.
197. Rosch RE, Wright S, Cooray G, Papadopoulou M, Goyal S, Lim M, et al. NMDA-
receptor antibodies alter cortical microcircuit dynamics. Proc Natl Acad Sci USA.
2018;115:E9916–25.
198. Adams NE, Hughes LE, Phillips HN, Shaw AD, Murley AG, Nesbitt D, et al. GABA-
ergic Dynamics in Human Frontotemporal Networks Confirmed by Pharmaco-
Magnetoencephalography. J Neurosci. 2020;40:1640–9.
199. Adams NE, Hughes LE, Rouse MA, Phillips HN, Shaw AD, Murley AG, et al.
GABAergic cortical network physiology in frontotemporal lobar degeneration.
Brain. 2021;144:2135–45.
200. Kaufman J, Gelernter J, Hudziak JJ, Tyrka AR, Coplan JD. The Research Domain
Criteria (RDoC) Project and Studies of Risk and Resilience in Maltreated Children.
J Am Acad Child Adolesc Psychiatry. 2015;54:617–25.
201. Wald A. An Essentially Complete Class of Admissible Decision Functions. Ann
Math Stat. 1947;18:549–55.
202. Brown LD. A Complete Class Theorem for Statistical Problems with Finite-
Sample Spaces. Ann Stat. 1981;9:1289–300.
203. Bowers JS, Davis CJ. Bayesian just-so stories in psychology and neuroscience.
Psychol Bull. 2012;138:389–414.
204. Schwartenbeck P, Friston K. Computational Phenotyping in Psychiatry: A
Worked Example. eNeuro. 2016;3:0049-16.2016.
205. Smith R, Kirlic N, Stewart JL, Touthang J, Kuplicki R, McDermott TJ, et al. Long-
term stability of computational parameters during approach-avoidance conflict
in a transdiagnostic psychiatric patient sample. Sci Rep. 2021;11:11783.
206. Smith R, Kirlic N, Stewart JL, Touthang J, Kuplicki R, Khalsa SS, et al. Greater
decision uncertainty characterizes a transdiagnostic patient sample during
approach-avoidance conflict: a computational modelling approach. J Psychiatry
Neurosci. 2021;46:E74–87.
Molecular Psychiatry (2023) 28:256 – 268
K. Friston
267
207. Iglesias S, Mathys C, Brodersen KH, Kasper L, Piccirelli M, den Ouden HEM, et al.
Hierarchical prediction errors in midbrain and basal forebrain during sensory
learning. Neuron. 2013;80:519–30.
208. Friston KJ, Redish AD, Gordon JA. Computational Nosology and Precision Psy-
chiatry. Comput Psychiatr. 2017;1:2–23.
209. Tervo DG, Tenenbaum JB, Gershman SJ. Toward the neural implementation of
structure learning. Curr Opin Neurobiol. 2016;37:99–105.
210. Wu JT, Leung K, Leung GM. Nowcasting and forecasting the potential domestic
and international spread of the 2019-nCoV outbreak originating in Wuhan,
China: a modelling study. Lancet (Lond, Engl). 2020;395:689–97.
211. Friston K, Costello A, Pillay D. ‘Dark matter’, second waves and epidemiological
modelling. BMJ Glob Health. 2020;5:e003978.
212. Testolin A, Zorzi M. Probabilistic Models and Generative Neural Networks:
Towards an Unified Framework for Modeling Normal and Impaired Neurocog-
nitive Functions. Front Computational Neurosci. 2016;10:73.
213. Parr T, Rees G, Friston KJ. Computational Neuropsychology and Bayesian
Inference. Front Hum Neurosci. 2018;12:61.
214. Smith R, Khalsa SS, Paulus MP. An Active Inference Approach to Dissecting
Reasons for Nonadherence to Antidepressants. Biol Psychiatry-Cogn Neurosci
Neuroimaging. 2021;6:919–34.
215. Sajid N, Parr T, Gajardo-Vidal A, Price CJ, Friston KJ. Paradoxical lesions, plasticity
and active inference. Brain Commun. 2020;2:fcaa164.
216. Proix T, Jirsa VK, Bartolomei F, Guye M, Truccolo W. Predicting the spatio-
temporal diversity of seizure propagation and termination in human focal
epilepsy. Nat Commun. 2018;9:1088.
217. Pfeffer CK, Xue M, He M, Huang ZJ, Scanziani M. Inhibition of inhibition in visual
cortex: the logic of connections between molecularly distinct interneurons. Nat
Neurosci. 2013;16:1068–76.
218. Yu AJ, Dayan P. Acetylcholine in cortical inference. Neural Netw. 2002;15:719–30.
219. Dayan P, Yu AJ. ACh, uncertainty, and cortical inference. In: Dietterich T, Becker
S, Ghahramani Z (eds). Advances in Neural Information Processing Systems. Vol
14, MIT Press; 2001.
220. Dayan P, Yu AJ. Phasic norepinephrine: a neural interrupt signal for unexpected
events. Netw: Comput Neural Syst. 2006;17:335–50.
221. Friston K, Schwartenbeck P, FitzGerald T, Moutoussis M, Behrens T, Dolan RJ. The
anatomy of choice: dopamine and decision-making. Philos Trans R Soc B: Biol
Sci. 2014;369:20130481.
ACKNOWLEDGEMENTS
I would like to thank Rick Adams, Thomas Parr, Richard Rosch, Christian Lambert, and
James Rowe for guidance on the presentation of this viewpoint.
AUTHOR CONTRIBUTIONS
All contributions were from the single author.
FUNDING
KF is supported by funding for the Wellcome Centre for Human Neuroimaging (Ref:
205103/Z/16/Z) and a Canada-UK Artificial Intelligence Initiative (Ref: ES/T01279X/1).
COMPETING INTERESTS
The author declares no competing interests.
GLOSSARY
Active inference securing evidence for a generative model of the world
through Bayesian inference and active sampling of
(sensory) data. This is equivalent to resolving uncertainty
and maximising model evidence.
Bayesian belief a posterior probability distribution over a random
variable, such as a latent cause or hidden state of the
world causing (sensory) data.
Belief updating the process of statistical inference, in which Bayes’
theorem is used to update a prior belief to a posterior
belief on the basis of new evidence.
Disconnection a term for neuropsychiatric syndromes caused by
(syndrome) damage to the white matter axons of associational or
commissural pathways: i.e., a disrupted connection.
Dysconnection a term for neuropsychiatric syndromes caused by a
(syndrome) synaptopathy involving associational or commissural
pathways: i.e., a dysfunctional connection.
 K. Friston
268
Inference updating beliefs to maximise model evidence. Varia- ADDITIONAL INFORMATION
tional Bayesian inference corresponds to minimising Correspondence and requests for materials should be addressed to Karl Friston.
variational free energy, which provides an upper bound
on log evidence. Reprints and permission information is available at http://www.nature.com/
Interoceptive pertaining to internal (e.g., autonomic) states reprints
Likelihood the probability of observing some (sensory) data, given
the causes of those data Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims
Model evidence the probability of some (sensory) data under a generative in published maps and institutional affiliations.
model. Also known as the marginal likelihood, i.e., having
marginalised over the causes of the data.
Precision The inverse variance of a random variable. More
generally, the precision of a probability distribution Open Access This article is licensed under a Creative Commons
decreases with its entropy. Attribution 4.0 International License, which permits use, sharing,
Prediction the prediction of some (sensory) data, based upon adaptation, distribution and reproduction in any medium or format, as long as you give
posterior beliefs about how those data were generated. appropriate credit to the original author(s) and the source, provide a link to the Creative
Prediction error A quantity used in predictive coding to denote the Commons license, and indicate if changes were made. The images or other third party
difference between an observation or point estimate and material in this article are included in the article’s Creative Commons license, unless
its predicted value. Predictive coding uses precision indicated otherwise in a credit line to the material. If material is not included in the
weighted prediction errors to update posterior beliefs article’s Creative Commons license and your intended use is not permitted by statutory
that generate predictions. regulation or exceeds the permitted use, you will need to obtain permission directly
Prior belief a belief prior to sampling (sensory) data. from the copyright holder. To view a copy of this license, visit http://
Sejunction (from the Latin noun seiunctio: divorce, separation) was creativecommons.org/licenses/by/4.0/.
hypothesis introduced by Carl Wernicke (1848-1904) to suggest that
psychopathology results from interruption (“sejunction”)
of associative connections in the brain. © The Author(s) 2022
Variational a tractable form of Bayesian belief updating based upon
inference minimising variational free energy (a.k.a., an evidence
bound). The gradients of variational free energy corre-
spond to precision weighted prediction errors.

Molecular Psychiatry (2023) 28:256 – 268